U.S. patent application number 15/677916 was filed with the patent office on 2019-02-21 for systems and methods for provision of a guaranteed batch.
This patent application is currently assigned to Google LLC. The applicant listed for this patent is Google LLC. Invention is credited to Ramy Abdelaal, Alexandre Duarte, Walfredo Cirne Filho, Maya Haridasan, Smeeta Jalan, Yingchong Situ, Robert van Gent.
Application Number | 20190058669 15/677916 |
Document ID | / |
Family ID | 62598088 |
Filed Date | 2019-02-21 |
![](/patent/app/20190058669/US20190058669A1-20190221-D00000.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00001.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00002.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00003.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00004.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00005.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00006.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00007.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00008.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00009.png)
![](/patent/app/20190058669/US20190058669A1-20190221-D00010.png)
View All Diagrams
United States Patent
Application |
20190058669 |
Kind Code |
A1 |
Duarte; Alexandre ; et
al. |
February 21, 2019 |
SYSTEMS AND METHODS FOR PROVISION OF A GUARANTEED BATCH
Abstract
Systems and methods for providing a guaranteed batch pool are
described, including receiving a job request for execution on the
pool of resources; determining an amount of time to be utilized for
executing the job request based on available resources from the
pool of resources and historical resource usage of the pool of
resources; determining a resource allocation from the pool of
resources, wherein the resource allocation spreads the job request
over the amount of time; determining that the job request is
capable of being executed for the amount of time; and executing the
job request over the amount of time, according to the resource
allocation.
Inventors: |
Duarte; Alexandre;
(Milpitas, CA) ; Situ; Yingchong; (Campbell,
CA) ; van Gent; Robert; (Redwood City, CA) ;
Filho; Walfredo Cirne; (Palo Alto, CA) ; Abdelaal;
Ramy; (Waterloo, CA) ; Jalan; Smeeta; (San
Jose, CA) ; Haridasan; Maya; (San Jose, CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Google LLC |
Mountain View |
CA |
US |
|
|
Assignee: |
Google LLC
Mountain View
CA
|
Family ID: |
62598088 |
Appl. No.: |
15/677916 |
Filed: |
August 15, 2017 |
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
H04L 41/5009 20130101;
H04L 47/78 20130101; H04L 41/5038 20130101; H04L 41/147 20130101;
G06F 9/5027 20130101; G06F 9/5005 20130101; G06F 9/5061 20130101;
H04L 41/14 20130101; H04L 41/50 20130101 |
International
Class: |
H04L 12/911 20060101
H04L012/911; G06F 9/50 20060101 G06F009/50 |
Claims
1. A computer-implemented method for managing a pool of resources,
the method comprising: receiving a job request for execution on the
pool of resources; determining an amount of time to be utilized for
executing the job request based on available resources from the
pool of resources and historical resource usage of the pool of
resources; determining a resource allocation from the pool of
resources, wherein the resource allocation spreads the job request
over the amount of time; determining that the job request is
capable of being executed for the amount of time; and executing the
job request over the amount of time, according to the resource
allocation.
2. The computer-implemented method of claim 1, wherein the
determining the amount of time to be utilized for executing the job
request comprises, for each unit of time: determining available
resources for the job request from the pool of resources based on
available units of resource time, and determining a period of time
that is capable of being allocated for the job request based on a
budget of resources allocated for the job request and the available
units of resource time.
3. The computer-implemented method of claim 2, further comprising
determining the available units of resource time based on available
resources from the pool of resources and estimated resource
commitments over one or more units of time based on historical
resource usage of the pool of resources.
4. The computer-implemented method of claim 3, wherein the
determining the available units of resource time comprises:
determining a first available units of resource time based on
available resources from the pool of resources and the estimated
resource commitments over one or more units of time based on
historical resource usage of the pool of resources; determining a
second available units of resource time based on available units of
resource time of a user associated with the job request, the
available units of resource time of the user associated with the
job request determined based on a resource allocation budget
assigned to the user associated with the job request and resource
commitments of the user associated with the job request.
5. The computer-implemented method of claim 3, wherein the
determining the period of time that is capable of being allocated
for the job request based on a budget of resources allocated for
the job request and the available units of resource time comprises:
for each unit of time, determining the resource allocation from the
pool of resources based on the available resources for the each
unit of time and the estimated commitments for the each unit of
time until the job request is fully allocated; and determining the
period of time from the resource allocation.
6. The computer-implemented method of claim 5, wherein for each
unit of time, the determining the resource allocation from the pool
of resources based on the available resources for the each unit of
time and the estimated commitments for the each unit of time until
the job request is fully allocated further comprises: for the each
unit of time, determining the resource allocation based on the
resource commitments of the user associated with the job request
for the each unit of time and a resource allocation budget assigned
to the user associated with the job request for the each unit of
time.
7. The computer-implemented method of claim 1, wherein the
determination that the job request is capable of being executed for
the amount of time is based on a determination that the amount of
time to be utilized for executing the job request meets a time
requirement associated with the job request.
8. The computer-implemented method of claim 1, wherein the
determination that the job request is capable of being executed for
the amount of time is based on a determination that a downtime
associated with the resource allocation does not exceed a maximum
downtime associated with the job request.
9. The computer-implemented method of claim 8, wherein the
determination that a downtime associated with the resource
allocation does not exceed the maximum downtime associated with the
job request comprises: determining the downtime of the resource
allocation based on determining one or more units of time in the
resource allocation being indicative of non-allocation; for the
downtime not exceeding the maximum downtime, determining that the
job request is capable of being executed; and for the downtime
exceeding the maximum downtime, determining that the job request is
not capable of being executed.
10. The computer-implemented method of claim 1, wherein the
determination that the job request is capable of being executed for
the amount of time is based on a determination that the resource
allocation can be executed.
11. The computer-implemented method of claim 1, wherein the
determining the resource allocation comprises, for each unit of
time: determining the resource allocation for the each unit of time
based on a budget associated with the job request and available
resources from the pool of resources for the each period of
time.
12. The computer-implemented method of claim 1, wherein the
executing the job request comprises, for each unit of time in the
amount of time, allocating resources from the pool of resources
according to the resource allocation corresponding to the each unit
of time.
13. The computer-implemented method of claim 1, further comprising:
managing a resource budget for each user of a plurality users of
the pool of resources; determining queuing for the plurality of
users based on resource commitments of the plurality of users and
the resource budget for the each user of the plurality of
users.
14. The computer-implemented method of claim 13, wherein the
determining an amount of time to be utilized for executing the job
request based on available resources from the pool of resources and
historical resource usage of the pool of resources comprises:
estimating resource commitments for each of the plurality of users
based on the historical resource usage of the pool, and estimating
the available resources from the pool of resources from the
estimated resource commitments.
15. The computer-implemented method of claim 14, wherein the
determining the resource allocation from the pool of resources
comprises, for each unit of time, determining the resource
allocation for the job request based on the estimated available
resources for the each unit of time.
16. The computer-implemented method of claim 13, wherein the pool
of resources comprises storage and central processing units (CPUs)
provided from a cloud pool, wherein the job request comprises a
request of at least one of processing and memory.
17. The computer-implemented method of claim 13, wherein the
managing the resource budget for each user of a plurality users of
the pool of resources comprises managing resource allocation to one
or more client devices associated with each user based on the
resource budget associated with each user.
18. The computer-implemented method of claim 13, wherein the
determining the resource allocation from the pool of resources is
based on the queuing of the user from the plurality of the users
associated with the job request.
19. The computer-implemented method of claim 18, wherein the
determining the resource allocation from the pool of resources is
based on the queuing of the user from the plurality of the users
associated with the job request comprises: for each unit of time,
estimating the queuing of the user associated with the job request;
and estimating resource commitments of users from the plurality of
users ahead of the user associated with the job request in the
queue.
20. A system configured to manage a pool of resources, the system
comprising: a processor, configured to: receive a job request from
the pool of resources; determine an amount of time to be utilized
for executing the job request based on available resources from the
pool of resources and historical resource usage of the pool of
resources; determine a resource allocation from the pool of
resources, wherein the resource allocation spreads the job request
over the amount of time; determine that the job request is capable
of being executed for the amount of time; and execute the job
request over the amount of time, according to the resource
allocation.
Description
BACKGROUND
Field
[0001] The subject matter of the present disclosure relates
generally to systems and methods for resource allocation and
management, and more particularly, to system and methods for
provision of a guaranteed batch.
Related Background
[0002] There are several related art implementations for scheduling
optimizations for distributed computing systems. For example,
related art implementations provide algorithms for optimizing
scheduling to improve fairness, user satisfaction, and utilization.
Related art implementations can also utilize deep learning to
improve the packing of tasks with multiple resource
requirements.
[0003] Related art budget based scheduling implementations can
involve schedulers for workflows with deadline and/or budget
guarantees using genetic algorithms. Other related art
implementations may attempt to allocate resources based on the user
Quality of Service (QoS) constraints, such as deadline and
budget.
[0004] Related art implementations have also involved dynamic cloud
implementations. In an example related art implementation, there is
a service level agreement (SLA)-aware platform for resource
management. There are related art implementations that involve a
learning system that can make explicit Service Level Objectives
(SLOs) based on historical data of periodic jobs and enforce such
SLOs using scheduling algorithms.
SUMMARY
[0005] Example implementations described herein are directed to
systems and methods to allocate resources for an intermittent load.
The example implementations may involve a per user budget-based
resource allocation that can obviate the need to provide deadline
or load presence information on a per job basis.
[0006] The subject matter includes a computer-implemented method
for managing a pool of resources, which can involve receiving a job
request for execution on the pool of resources; determining an
amount of time to be utilized for executing the job request based
on available resources from the pool of resources and historical
resource usage of the pool of resources; determining a resource
allocation from the pool of resources, wherein the resource
allocation spreads the job request over the amount of time;
determining that the job request is capable of being executed for
the amount of time; and executing the job request over the amount
of time, according to the resource allocation.
[0007] The subject matter further includes a non-transitory
computer readable medium, storing instructions for managing a pool
of resources, the instructions which can involve receiving a job
request for execution on the pool of resources; determining an
amount of time to be utilized for executing the job request based
on available resources from the pool of resources and historical
resource usage of the pool of resources; determining a resource
allocation from the pool of resources, wherein the resource
allocation spreads the job request over the amount of time;
determining that the job request is capable of being executed for
the amount of time; and executing the job request over the amount
of time, according to the resource allocation.
[0008] The subject matter further includes a system configured to
manage a pool of resources, the system which can involve a
processor, configured to receive a job request from the pool of
resources; determine an amount of time to be utilized for executing
the job request based on available resources from the pool of
resources and historical resource usage of the pool of resources;
determine a resource allocation from the pool of resources, wherein
the resource allocation spreads the job request over the amount of
time; determine that the job request is capable of being executed
for the amount of time; and execute the job request over the amount
of time, according to the resource allocation.
[0009] The subject matter also includes a system for managing a
pool of resources, the system which can involve means for receiving
a job request for execution on the pool of resources; means for
determining an amount of time to be utilized for executing the job
request based on available resources from the pool of resources and
historical resource usage of the pool of resources; means for
determining a resource allocation from the pool of resources,
wherein the resource allocation spreads the job request over the
amount of time; means for determining that the job request is
capable of being executed for the amount of time; and means for
executing the job request over the amount of time, according to the
resource allocation.
[0010] The methods are implemented using one or more computing
devices and/or systems. The methods may be stored in
computer-readable media such as a non-transitory computer readable
medium.
BRIEF DESCRIPTION OF THE DRAWINGS
[0011] FIGS. 1(a) and 1(b) illustrate examples of resource
multiplexing in accordance with an example implementation.
[0012] FIG. 2 illustrates an example cloud environment, in
accordance with an example implementation.
[0013] FIGS. 3(a) to 3(c) illustrate example management information
that can be utilized by the management system, in accordance with
an example implementation.
[0014] FIGS. 4(a) to 4(d) illustrate examples of process
implementations that can be executed by the management system, in
accordance with an example implementation.
[0015] FIG. 5 shows an example environment suitable for some
example implementations.
[0016] FIG. 6 shows an example computing environment with an
example computing device suitable for use in some example
implementations.
DETAILED DESCRIPTION
[0017] The subject matter described herein is taught by way of
example implementations. Various details have been omitted for the
sake of clarity and to avoid obscuring the subject matter. The
examples shown below are directed to structures and functions for
implementing systems and methods for provision of a guaranteed
batch. Terms used throughout the description are provided as
examples and are not intended to be limiting.
[0018] For example, the use of the term "automatic" may involve
fully automatic or semi-automatic implementations involving user or
administrator control over certain aspects of the implementation,
depending on the desired implementation of one of ordinary skill in
the art practicing implementations of the present application.
Selection can be conducted by a user through a user interface or
other input means, or can be implemented through a desired
algorithm. Example implementations as described herein can be
utilized either singularly or in combination and the functionality
of the example implementations can be implemented through any means
according to the desired implementations.
[0019] Cloud infrastructure workloads can often be broadly
classified, in terms of their presence in the system, as either
continuous or intermittent. Continuous workloads involve workloads
that are always present in the system and can include both
user-facing (e.g., latency-sensitive) server applications as well
as continuous background (e.g., latency-tolerant) data-processing
systems (e.g., data compression, web crawling/indexing).
[0020] Intermittent workloads are workloads that are characterized
by varying periods of absence intercalated by periods of activity,
where resources are actually required and/or consumed. This group
mostly encompasses intermittent background data-processing systems
(e.g., logs processing, simulations).
[0021] Load presence can be important for resource management
(e.g., efficiency). Continuous workloads require continuous
resource allocation, consuming a certain (usually varying over a
determined interval) amount of resources for 24 hours/day.
Variability in the consumption may lead to waste, as resources must
be provisioned for the peak load. Cloud providers rely on different
availability Service Level Objectives (SLO) to meet the user
requirements without hurting overall fleet efficiency.
[0022] Intermittent workloads can be seen as a special class of
continuous workloads with extreme consumption variability
conditions (from 0 to peak). Therefore, continuous resource
allocation can also be used for intermittent workloads, but such
allocations can be very wasteful for intermittent workloads.
Furthermore, the availability SLOs associated with continuous
resource allocations may not be as meaningful for intermittent
workloads as they are to continuous workloads, especially when the
ratios between presence periods and absence periods approach
zero.
[0023] Intermittent resource allocation, where resources are
reserved for a given user for a pre-defined period and freed for
the remaining time, is the solution to avoid increasing resource
idleness time. However, intermittent resource allocation is much
harder to achieve than continuous resource allocation, because it
puts an extra burden on the users, requiring the users to specify
not only the required amount of resources, but also when and for
how long these resources will actually be needed, and on the
provider, that must keep a complex scheduling of multiple
intermittent resource allocations to be able decide on the
feasibility of new user requests.
[0024] In related art implementations, large cloud infrastructures
address intermittent load by using reclaimed resources. In these
related art cases, the intermittently available resources are
usually much cheaper than their continuous counterparts, in
exchange for providing none or, at least, very weak SLOs.
[0025] Example implementations described herein are directed to a
resource management model to allocate resources for intermittent
load, or batch load, that relies on per user budgets and which may
obviate the need for information regarding job duration or
deadlines. Example implementations as described herein utilize a
novel definition of a user budget, representing the maximum amount
of resources a user can consume over a fixed period of time (e.g.,
constant 24 hours, but not limited thereto), hence, defined using a
RESOURCE*TIME unit (e.g., CPU*HOURS, BYTES*HOUR, and so on).
[0026] Example implementations described herein can be applied to
cloud providers and other entities (e.g., enterprises such as
companies) with large distributed computing infrastructures
attending multiple individual or groups of users to improve
utilization/efficiency on the computing clusters.
[0027] FIGS. 1(a) and 1(b) illustrate an example of resource
multiplexing in accordance with an example implementation. As
illustrated in FIG. 1(a), example implementations described herein
time multiplex resources by transforming the shape of jobs to fit
in the capacity of the system (e.g., spread the job request over
the amount of time). Each job request may include information such
as a list of tasks to be completed by the job request, a requested
resource allocation, and one or more constraints on the job
request. Each job may involve an aggregation of individual tasks
(e.g., execute program A, conduct process B, feed result into
process C, etc.), which may be run consecutively or concurrently,
and which may be prioritized.
[0028] There is a high probability that at peak times, not all
admitted jobs will fit into the capacity of the pool. To handle
such situations, example implementations may disable individual
tasks within jobs as needed to ensure that the SLOs of all admitted
jobs are met, as shown in FIG. 3(c) as discussed below. Disabled
tasks can be marked on the user interface (UI) as "Disabled Due To
Congestion," and such terminations will not count towards user
configured failure limits for the job/task. In example
implementations, jobs running in `best effort` mode will be sized
down or disabled completely, as necessary. If the system is still
congested after such activity, jobs with guarantees which are
farther away from their expected deadline (e.g., creation time+24
hours) will be disabled (e.g., squeezed) before the ones which are
closer to make sure that no guarantees on job completion would be
violated.
[0029] In the example of FIG. 1(b), as users are assigned a
guaranteed batch in terms of RESOURCE*TIME, the ceiling/commitment
of the user is considered as an average throughput over a period of
time (e.g., 24 hours). Accordingly, a commitment of Y resources
means that the user may be able to consume, on average, Y units of
this resource over the next period of time. The guaranteed batch
workload may exceed Y when more resources are available and may be
squeezed during contention. For example, if a user has a commitment
of ten resources (e.g., Central Processing Units (CPUs), memory
units, etc.) and is using all of the resources for a guaranteed
batch, examples of the possible job consumption patterns are shown
for FIG. 1(b). Note that the resource usages over time (area of
shaded area) are exactly the same in the three illustrated
cases.
[0030] FIG. 2 illustrates an example cloud environment 200, in
accordance with an example implementation. In an example
implementation, the cloud environment 200 can include one or more
client devices 205, a network 210, one or more resource batch pools
215 (e.g., one or more servers) and a management system 220. As
used herein, "pools" may refer to one or more pools of resources,
which may include (but are not limited to) one or more apparatuses
or systems that may be configured to provide processing power with
respect to one or more of CPUs, storage from storage systems in
terms of flash memory or disk, specialized GPUs, specialized
programs, or the like, depending on the desired implementation. The
one or more client devices 205 execute one or more jobs on
resources allocated from the one or more resource batch pools 215,
and may access such resources through network 210. Examples of such
client devices 205 are provided with respect to FIG. 5. Network 210
may be configured to facilitate connections between the one or more
client devices 205, the one or more resource batch pools 215 and
the management system 220. Such a network 210 may be in the form of
a Local Area Network (LAN), a Wide Area Network (WAN), or otherwise
depending on the desired implementation. The one or more resource
batch pools 215 may be implemented in the form of a cloud including
any combination of dedicated central processing units (CPUs),
storage systems, memory units, general processing units (GPUs),
servers, and other hardware units that are configured to provide
desired resources for the one or more client devices 205. Resources
that can be provided can include, but are not limited to, CPUs,
memory, bandwidth, specialized processors, and so on according to
the desired implementation.
[0031] In example implementations, the one or more resource batch
pools 215 may be managed by a management system 220. Management
system 220 can be configured to receive requests for a resource
allocation for a job from the one or more client devices 205, and
then manage the allocation of resources from the resource batch
pools 215 for the one or more clients 205 based on the example
implementations described herein, and as shown in FIG. 3(c) and
further discussed below. For example, a resource allocation may
provide an indication of resources that are required to execute the
job associated with the job request, and may be specified in terms
of RESOURCE*TIME (e.g., CPU*TIME, MEMORY*TIME, etc.), and/or may be
specified in terms of resources (e.g., CPU, MEMORY).
[0032] In example implementations, a batch pool is utilized as an
entity holding the resources available for users with batch load as
illustrated by the resource batch pool 215 of FIG. 2, and provides
a pool of resources for use. The capacity from a given batch pool
is obtained by blocking computing resources for a specified amount
of time .DELTA.t, hence defined as RESOURCE*HOUR/.DELTA.t. Such a
batch pool provides different products for the pool of resources
(e.g., CPU, memory, disk, etc.). For example, the pool of resources
may include one or more apparatuses or systems configured to
provide any combination of processing power in terms of CPUs,
storage from storage systems in terms of flash memory or disk,
specialized GPUs, specialized programs, and so on depending on the
desired implementation. The capacity of each batch product p,
bc.sub.p, expressed in RESOURCE*HOUR/period of time .DELTA.t units,
is defined in terms of the capacity allocated from the underlying
continuous product for the period .DELTA.t, cc.sub.p, as:
bc.sub.p=(cc.sub.p*.DELTA.t) (1)
[0033] For example, a pool administrator can fund a batch pool by
allocating 1000 RESOURCE units of a continuous product p, over a
period of time (e.g., 24 hours). The resulting capacity of the
equivalent batch product would be:
bc.sub.p=(1000*24) or 24000 RESOURCE*HOURs/DAY
[0034] The batch pool capacity can be defined as the integral of
the instantaneous capacity allocated from the underlying continuous
product p at any moment t, cc.sub.p(t), over a period of time
.DELTA.t:
bc p = .intg. t - .DELTA. t t cc p ( t ) dt ( 2 ) ##EQU00001##
[0035] Using this definition the 24000 RESOURCE*HOURs/DAY, batch
capacity mentioned in the previous example could be obtained both
by allocating 1000 RESOURCE units for 24 hours or 2000 RESOURCE
units for 12 hours over a period of 24 hours, for example.
[0036] Similarly, the fraction of the batch pool capacity allocated
for a particular user, the user commitments, over a period of time
.DELTA.t can be expressed as:
bc up = .intg. t - .DELTA. t t cc up ( t ) dt ( 3 )
##EQU00002##
[0037] In the example implementations, users initially start with
the same commitments (e.g., 0) and have the number adjusted
periodically according to their past consumption. Such a definition
does not imply any particular bound on the instantaneous capacity
of the underlying continuous product at any moment t.
[0038] Batch Pool capacity (bc.sub.p) and Batch User Commitments
(bc.sub.up) are concepts valid over a period of time .DELTA.t, and
no assumptions are made about the underlying instantaneous values
at any particular instant t.
[0039] In example implementations, the proposed resource allocation
mode can provide two SLOs for users with intermittent load. In a
first SLO, users with sufficient load will be able to consume their
entire budget (in this case, commitments) over a period .DELTA.t.
In a second SLO, the downtime for any job j, dj, over a period of
time .DELTA.t does not exceed the maximum specified downtime D. A
job can be considered as experiencing downtime if a particular task
in the job request cannot access the minimum required resources to
execute the task, when overall resource allocation for the job
falls below a certain threshold, or can be defined by other methods
depending on the desired implementation.
[0040] The submitted load, sl, is defined as all of the load a user
submits to a batch pool and admitted load, al, is defined as the
amount of the sl admitted by the batch scheduler to run under an
SLO. In an example implementation and as shown in FIG. 3(c) and
discussed below, users can submit jobs that specify the required
resources to start running the job and the maximum downtime D the
job can take. In such an example, the first SLO can be utilized by
enforcing the following two inequalities during resource
allocation:
.A-inverted. t , .intg. t - .DELTA. t t al pu ( t ) dt .gtoreq. MIN
( bc pu , .intg. t - .DELTA. t t sl pu ( t ) dt ) ( 4 )
.A-inverted. t , al pu ( t ) .ltoreq. sl pu ( t ) ( 5 )
##EQU00003##
[0041] Inequality (4) specifies that for every moment t, the
admitted load for user u over the last .DELTA.t period of time is
the minimum of the commitments and the submitted load. That is, if
(submitted load.ltoreq.commitments), all submitted load is admitted
by the system. On the contrary, if commitments<submitted load,
only the load amounting to the commitments will be accepted to
receive the SLO.
[0042] Inequality (5) is a control mechanism to prevent allocating
more resources than the user can actually consume at any moment t.
The second SLO imposes a constraint on how long jobs can be delayed
and can be obtained with job prioritization according to the
remaining user budget over .A-inverted.t.
[0043] Based on the equations above, the criteria C(u) to identify
the user u to have the next job released is:
C = MAX ( 0 , bc pu - .intg. t - .DELTA. t t cc pu ( t ) dt )
##EQU00004##
[0044] The user with the greatest C(u) is ahead of the queue. For a
given user, the jobs are dequeued in first in first out (FIFO)
order.
[0045] FIG. 3(a) illustrates example management information as
utilized by the management system, in accordance with an example
implementation. Specifically, FIG. 3(a) illustrates example
management information for management of resource allocation for
one or more users. The management information of FIG. 3(a)
facilitates the management system 220 to manage the resource budget
for each user of a plurality users of the pool of resources through
managing resource allocation to one or more client devices
associated with each user based on the resource budget associated
with each user. As the client devices may involve multiple
different users, or as a user may utilize multiple client devices
to execute various jobs, resource budgets can be managed at a user
level as described below.
[0046] The management information as illustrated in FIG. 3(a) can
include user identifiers (IDs) 300, an associated resource
allocation budget 301, present resource commitments 302, historical
resource usage 303, and pending job requests 304 for resource
allocation. The user ID 300 can indicate the identification of the
user that has job requests running on the resource batch pools 215.
The resource allocation budget 301 can be represented as a vector
across all of the different types of resources available in the
resource batch pools 215, with each element in the vector
indicative of the amount of available resources for the particular
type of resource. Such resources can be represented as
RESOURCE*TIME, with the time based on the time period associated
with the type of resource, and calculated based on equation
(3).
[0047] The present resource commitments can include the jobs
presently executed by the user and the resources consumed, as
illustrated in FIG. 1(b). The present resource commitments 302 may
also include a vector across all of the different types of
resources available in the resource batch pools 215, with each
element in the vector indicative of the amount of resources
consumed for the particular type of resource.
[0048] Such values can be determined from the summation of the
areas of the resources utilized over time for each type of resource
for each job as illustrated in FIG. 1(b). Historical resource usage
303 can include the jobs historically executed by the user and the
associated historical resource usage or consumption as illustrated
in FIG. 1(b). Once jobs are completed for the user, the job can be
moved from the present resource commitments 302 to the historical
resource usage 303.
[0049] Job requests 304 can include pending job requests for
resource allocation submitted by the client device associated with
the user ID. The job requests 304 may eventually be processed in
accordance with the flow of FIG. 4(a), as explained below.
[0050] FIG. 3(b) illustrates example management information for
management of the pool of resources for the one or more resource
batch pools 215, in accordance with an example implementation. Such
management information can include types of resources 310, time
period for each type of resource 311, current resource commitments
for the type of resource 312, historical resource usage 313, and
expected resource commitments 314. Types of resources 310 can
include the type of resource available for resource batch pools
215, such as storage and central processing units (CPUs) provided
from a cloud pool, GPUs, bandwidth, or other resources depending on
the desired implementation, as well as the total resources for the
type of resource in the resource batch pool 215 (e.g., the batch
pool capacity for the type of resource calculated from equation
(2)). The time period for each type of resource 311 can include the
set time period (e.g., 24 hours) for each of the resources, and the
unit of time associated with the resource type (e.g., 1 hour), to
define the resource availability in terms of RESOURCE*TIME. The
time period and unit of time can be set by an administrator or
though other methods in accordance with the desired
implementation.
[0051] Current resource commitments 312 can include the job
information as illustrated in FIG. 1(b) across all users of the
resource batch pool 215 that utilize the particular type of
resource. Current resource commitments 312 can also include a value
indicating how much of the particular type of resource is consumed
in terms of RESOURCE*TIME.
[0052] Historical resource usage 313 can include the job
information as illustrated in FIG. 1(b) from all of the jobs that
utilize the particular type of resource. When a job in the current
resources commitments is completed, such job information can be
moved to the historical resource usage 313.
[0053] Expected resource commitments 314 can include information as
illustrated in FIG. 1(b) that is calculated based on historical
resource usage 313. Such a calculation can be conducted by taking
an average of resources consumed for each time unit for a time
period, or through other methods, depending on the desired
implementation.
[0054] FIG. 3(c) illustrates an example breakdown of a job request,
in accordance with an example implementation. Each job request
submitted by a client device is associated with a job request ID
304-1 and can include information provided by the client device
such as a list of tasks 304-2 to be completed by the job request, a
requested resource allocation 304-3, and one or more constraints on
the job request 304-4. Each job may involve an aggregation of
individual tasks 304-2 (e.g., execute program A, conduct process B,
feed result into process C, etc.), which can run (e.g., at one or
more servers) consecutively or concurrently in any combination,
depending on the specifications of the job request. Tasks may also
be prioritized in terms of importance, so that users can prioritize
which tasks are to be disabled last, which tasks cannot be disabled
for the job request, and so on, depending on the desired
implementation.
[0055] Resource allocation 304-3 can indicate the required
resources to execute the job associated with the job request ID.
Resource allocation 304-3 can be specified in terms of
RESOURCE*TIME as described above (e.g., CPU*TIME, MEMORY*TIME),
and/or can be specified in terms of resources (e.g., CPU, MEMORY)
depending on the desired implementation. For example, but not by
way of limitation, the required resources may be provided at a
server, in response to the information provided by the client
device.
[0056] Constraints 304-4 can indicate constraints associated with
the job request ID, such as the maximum downtime the job or
particular tasks are allowed to have, time constraints in executing
the job request, specified SLOs and resource minimums for tasks,
and so on, depending on the desired implementation.
[0057] FIGS. 4(a) to 4(d) illustrate examples of process
implementations that can be executed by the management system, in
accordance with an example implementation.
[0058] FIG. 4(a) illustrates an example of a process implementation
400, in accordance with an example implementation. The process 400
of FIG. 4 illustrates an example for a process for the flow at FIG.
2. At 405, the process receives a job request for execution on the
pool of resources. Such a request may be received from the one or
more clients 205 by the management system 220. Such a job request
can involve a list of tasks, constraints for the job request, and a
request of at least one of processing, memory, bandwidth, GPUs or
specialized processors, and so on, depending on the desired
implementation, and received by the management system 220 to be
managed as illustrated in FIG. 3(c).
[0059] At 410, the process determines the amount of time utilized
for execution of the job request. The process can refer to the
available resources from the pool of resources and historical usage
data of the pool of resources from FIG. 3(b). In an example, the
available units of resource time can be determined for the resource
batch pool 215 based on the available resources from the pool of
resources (e.g., total resource units associated with a type of
resource from FIG. 3(b)) and the estimated resource commitments
over one or more units of time based on historical resource usage
313 of the pool of resources as indicated in estimated resource
commitments of FIG. 3(b).
[0060] Moreover, if a constraint in the job request requires that
the job be executed within a certain period of time, the time
provided in the constraint can be utilized as the amount of time
utilized for execution. Depending on the desired implementation,
time may also be a constant (e.g., 24 hours) in accordance to how
RESOURCE*TIME is determined in the desired implementation.
[0061] The available units of resource time can be determined from
the difference between the total resource units the estimated
resource commitments. Similarly, available units of resource time
for a user ID associated with the job request can be determined
based on a resource allocation budget assigned to the user
associated with the job request and resource commitments of the
user associated with the job request as indicated in FIG. 3(a). The
available units of resource time can be determined from the
difference between the budget and the resource commitments.
[0062] At 415, the process determines resource allocation from the
pool of resources based on the amount of time utilized for the job
request. Depending on the present resource commitments of the user
associated with the request and the expected resource commitments
314 over the time period, an allocation is determined as
illustrated in FIGS. 1(a) and 1(b) that spreads the job request
over the period of time.
[0063] As illustrated in FIGS. 1(a) and 1(b), the job request does
not have to be fulfilled in uniform allocations; any process can be
utilized to allocate any amount of resources for a particular
period of time to spread the job request over the period of time.
For example, the job request can be spread in a greedy manner,
wherein the job request is given all of the expected free resources
for a particular unit of time based on the historical usage, and
constrained by the RESOURCE*TIME budget of the associated user or
up to the resources needed to execute the job. The job request may
also be spread according to priority of other tasks determined to
be executed in that particular unit of time. Other operations for
spreading the job request may be substituted therefor, as would be
understood by those skilled in the art.
[0064] As illustrated in FIG. 4(c), the determining the resource
allocation can involve a process that for each unit of time,
determines the resource allocation for the each unit of time based
on a budget associated with the job request and available resources
from the pool of resources for the each period of time. As
illustrated in FIG. 4(b), the determining the resource allocation
from the pool of resources can also involve, for each unit of time,
determining the resource allocation for the job request based on
the estimated available resources for the each unit of time.
[0065] Depending on the desired implementation, queuing is also
applied for the resource allocation. The resource allocation from
the pool of resources is determined based on the queuing of the
user from the plurality of the users associated with the job
request, and the other expected commitments received from other
users, based on the proposed criteria C(u). In conjunction with the
example implementations described in FIGS. 4(b) and 4(c), the
resource allocation can be conducted for each unit of time while
taking queuing into account. For example, for each unit of time,
the process can estimate the queuing of the user associated with
the job request based on the proposed criteria C(u); and estimate
resource commitments of users from the plurality of users ahead of
the user associated with the job request in the queue.
[0066] At 420, the process determines whether the job request is
capable of being executed for the amount of time. That is, a check
is performed to determine if the proposed amount of time to execute
the job request meets with the SLO requirement of the job request
or the user as defined by the constraints 304-4 from FIG. 3(c),
and/or if the resource request is valid (e.g., can be executed over
the specified time period).
[0067] For example, but not by way of limitation, it may not be
possible to execute the job request within the time defined by the
constraint associated with a job request, whereupon the job may be
returned with a message indicative of the constraint not being
possible. Such a determination can be made by comparing the
available RESOURCE*TIME units of the user with the requested
RESOURCE*TIME units, and also by comparing the RESOURCE*TIME units
expected to be free during the time period defined by the
constraint with the requested RESOURCE*TIME units. If there is
insufficient RESOURCE*TIME units in the pool of resources, or the
user does not have the sufficient RESOURCE*TIME units for the given
time constraint, then the job is considered to be invalid.
[0068] It may also not be possible to execute the job request
within the time defined by the administrator of the management
system as the basis for determining RESOURCE*TIME (e.g., 24 hours),
in which case the job may be returned with a message indicative of
the job not being executable at that time, or that the user/pool of
resources does not have sufficient resources to execute the job
request.
[0069] At 425, should the process determine that the job request is
capable of being executed for the amount of time, the process
executes the job request over the amount of time according to the
resource allocation. The executing the job request can involve, for
each unit of time in the amount of time, allocating resources from
the pool of resources according to the determined resource
allocation corresponding to the each unit of time, as illustrated
in FIG. 1(a).
[0070] Other executions of the job request may also be possible,
depending on the desired implementation. For example, if the
predicted usage is more than the currently experienced resource
consumption, more resources may be allocated to the job request for
a unit of time according to the desired implementation (e.g.,
proportionally to the other job requests being executed, based on
priority of job requests currently executed, etc.).
[0071] If the predicted usage is less than the currently
experienced resource consumption, then job requests may be given
proportionately less resources based on priority, or by other
metrics according to the desired implementation. In such an example
implementation, the execution of the job request at 425 can involve
determining if the currently experienced resource consumption is
less than the predicted resource consumption derived from
historical usage. If so, then job requests executed during the unit
of time can be allocated more resources according to the desired
implementation. If not, then job requests executed during the unit
of time can be provided fewer resources in a manner according to
the desired implementation.
[0072] Depending on the desired implementation, queuing may be
employed for the received plurality of requests. In such an example
implementation, the process manages the resource budget for each of
the users as illustrated in FIG. 3(a), and determines the queuing
order for the job requests based on the user associated with the
job request. The queuing is determined based on resource
commitments of the plurality of users and the resource budget for
the each user of the plurality of users as described with respect
to the proposed criteria C(u).
[0073] FIG. 4(b) illustrates a process for determining the amount
of time to be utilized for executing the job request, in accordance
with an example implementation. The process begins by, for each
unit of time, determining the available units of resource time at
410-1. At 410-2, the process then determines available resources
for the job request from the pool of resources based on available
units of resource time from referring to management information of
FIGS. 3(a) and 3(b). Through reference to the management
information as illustrated in FIGS. 3(a) and 3(b), the available
resources from the pool of resources can be determined and the
process can estimate resource commitments over one or more units of
time based on historical resource usage 313 of the pool of
resources. For example, the resource commitments of the user can be
estimated based on the historical resource usage 313 of the
resource batch pool 215 as described with respect to the expected
resource commitments 314 of FIG. 3(b), and the process can estimate
the available resources from the pool of resources from the
estimated resource commitments, which can involve taking the
difference between the total resources available and the estimated
resource commitments.
[0074] At 410-3, the process then determines a period of time that
is capable of being allocated for the job request based on a budget
of resources allocated for the job request and the available units
of resource time.
[0075] FIG. 4(c) illustrates an example flow for determining the
period of time for a job request and conducting the resource
allocation at 415. Specifically, FIG. 4(c) illustrates an example
process for the flow of 410 and 415 for FIG. 4(c), as an example
flow for determining the period of time that is capable of being
allocated for the job request based on a budget of resources
allocated for the job request and the available units of resource
time comprises. At 411-1, a loop is initiated for each unit of
time, the unit of time defined by the time period set in the
management information of FIG. 3(b). At 411-2, the process
estimates the resource commitments for the unit of time. The
estimation can be referenced from the estimation of resource
commitments from the management information of FIG. 3(b).
[0076] At 411-3, a resource allocation is determined for the unit
of time. Such a resource allocation can be determined based on the
resource commitments of the user associated with the job request
for the each unit of time and a resource allocation budget assigned
to the user associated with the job request for the each unit of
time from referring to the present commitments of the user ID
associated with the job request, the budget for the user ID, the
total resources for the type of resource in the resource batch pool
215, and the current and expected commitments for that type of
resource from the management information of FIGS. 3(a) and
3(b).
[0077] The current and expected commitments for that type of
resource are subtracted from the total resources to determine the
expected available resources for the unit of time. Similarly, the
present user commitments are subtracted from the user budget to
determine the expected resources available to the user for that
time unit. If the user can consume the resource budget (e.g.,
expected resources available to the user<expected available
resources for the unit of time), then resources up to the expected
resources available to the user are allocated as indicated in
equations (4) and (5). Otherwise, the allocation is done based on
the SLO policy as indicated in equation (5).
[0078] At 411-5, after all of the allocation is completed, the
amount of time units utilized is determined from the resource
allocation as the period of time to be associated with the job
request, and the proposed resource allocation is provided.
[0079] FIG. 4(d) illustrates example process for evaluating whether
the determined period of time is acceptable. Such a process can be
utilized at the process of 420. In an example as to whether the
determined period of time is acceptable, one method is to determine
whether a downtime associated with the resource allocation does not
exceed the maximum downtime associated with the job request.
[0080] At 420-1, the process extracts one or more units of time
from the resource allocation indicative of non-allocation (e.g.,
when zero resources are allocated for that particular time unit).
At 420-2, the process sums up the one or more units of time that
are indicative of non-allocation to determine the estimated
downtime. At 420-3, a determination is made as to whether the
non-allocation time (estimated downtime) exceeds the maximum
downtime. For the downtime not exceeding the maximum downtime (No),
the process proceeds to 420-4 and determines that the resource
allocation can be executed in the amount of time, therefore capable
of being executed. Otherwise, for the downtime exceeding the
maximum downtime (Yes), the flow proceeds to 420-5 and determines
that the job request is not capable of being executed.
[0081] Other methods can be utilized in conjunction with, or
separately from FIG. 4(d) to determine if the proposed period of
time and resource allocation is acceptable. Such a determination
can be made based on whether or not the SLOs as described above are
met. The job request may be associated with a time requirement, so
the determination can be based on whether the amount of time to be
utilized for executing the job request meets a time requirement
associated with the job request. Another example for a
determination whether the job request is capable of being executed
for the amount of time can be based on whether the proposed
resource allocation can be executed (e.g., is invalid or not due to
violating constraints associated with present job commitments,
other users, and so on).
[0082] In some examples, process 400 may be implemented with
different, fewer, or more blocks. Process 400 may be implemented as
computer executable instructions, which can be stored on a medium,
loaded onto one or more processors of one or more computing
devices, and executed as a computer-implemented method.
[0083] FIG. 5 shows an example environment suitable for some
example implementations. Environment 500 includes devices 505-545,
and each is communicatively connected to at least one other device
via, for example, network 550 (e.g., by wired and/or wireless
connections). Some devices may be communicatively connected to one
or more storage devices 530 and 545. Such devices 505-545 can be
examples of client devices 205, or can also be a device utilized to
facilitate the functionality of the management system 220.
[0084] An example of one or more devices 505-545 may be computing
device 605 described below in FIG. 6. Devices 505-545 may include,
but are not limited to, a computer 505 (e.g., a laptop computing
device), a mobile device 510 (e.g., smartphone or tablet), a
television 515, a device associated with a vehicle 520, a server
computer 525, computing devices 535-540, storage devices 530 and
545.
[0085] FIG. 6 shows an example computing environment with an
example computing device suitable for use in some example
implementations, such as management system 220 or client devices
205. Computing device 605 in computing environment 600 can include
one or more processing units, cores, or processors 610, memory 615
(e.g., RAM, ROM, and/or the like), internal storage 620 (e.g.,
magnetic, optical, solid state storage, and/or organic), and/or I/O
interface 625, any of which can be coupled on a communication
mechanism or bus 630 for communicating information or embedded in
the computing device 605.
[0086] Computing device 605 can be communicatively coupled to
input/user interface 635 and output device/interface 640. Either
one or both of input/user interface 635 and output device/interface
640 can be a wired or wireless interface and can be detachable.
Input/user interface 635 may include any device, component, sensor,
or interface, physical or virtual, that can be used to provide
input (e.g., buttons, touch-screen interface, keyboard, a
pointing/cursor control, microphone, camera, braille, motion
sensor, optical reader, and/or the like). Output device/interface
640 may include a display, television, monitor, printer, speaker,
braille, or the like. In some example implementations, input/user
interface 635 and output device/interface 640 can be embedded with
or physically coupled to the computing device 605. In other example
implementations, other computing devices may function as or provide
the functions of input/user interface 635 and output
device/interface 640 for a computing device 605.
[0087] Examples of computing device 605 may include, but are not
limited to, highly mobile devices (e.g., smartphones, devices in
vehicles and other machines, devices carried by humans and animals,
and the like), mobile devices (e.g., tablets, notebooks, laptops,
personal computers, portable televisions, radios, and the like),
and devices not designed for mobility (e.g., desktop computers,
other computers, information kiosks, televisions with one or more
processors embedded therein and/or coupled thereto, radios, and the
like). [96] Computing device 605 can be communicatively coupled
(e.g., via I/O interface 625) to external storage 645 and network
650 for communicating with any number of networked components,
devices, and systems, including one or more computing devices of
the same or different configuration. Computing device 605 or any
connected computing device can be functioning as, providing
services of, or referred to as a server, client, thin server,
general machine, special-purpose machine, or another label.
[0088] The I/O interface 625 may include wireless communication
components (not shown) that facilitate wireless communication over
a voice and/or over a data network. The wireless communication
components may include an antenna system with one or more antennae,
a radio system, a baseband system, or any combination thereof.
Radio frequency (RF) signals may be transmitted and received over
the air by the antenna system under the management of the radio
system.
[0089] I/O interface 625 can include, but is not limited to, wired
and/or wireless interfaces using any communication or I/O protocols
or standards (e.g., Ethernet, 802.11x, Universal System Bus, WiMax,
modem, a cellular network protocol, and the like) for communicating
information to and/or from at least all the connected components,
devices, and network in computing environment 600. Network 650 can
be any network or combination of networks (e.g., the Internet,
local area network, wide area network, a telephonic network, a
cellular network, satellite network, and the like).
[0090] Computing device 605 can use and/or communicate using
computer-usable or computer-readable media, including transitory
media and non-transitory media. Transitory media include
transmission media (e.g., metal cables, fiber optics), signals,
carrier waves, and the like. Non-transitory media include magnetic
media (e.g., disks and tapes), optical media (e.g., CD ROM, digital
video disks, Blu-ray disks), solid state media (e.g., RAM, ROM,
flash memory, solid-state storage), and other non-volatile storage
or memory.
[0091] Computing device 605 can be used to implement techniques,
methods, applications, processes, or computer-executable
instructions in some example computing environments.
Computer-executable instructions can be retrieved from transitory
media, and stored on and retrieved from non-transitory media. The
executable instructions can originate from one or more of any
programming, scripting, and machine languages (e.g., C, C++, C#,
Java, Visual Basic, Python, Perl, JavaScript, and others).
[0092] Processor(s) 610 can execute under any operating system (OS)
(not shown), in a native or virtual environment. One or more
applications can be deployed that include logic unit 660,
application programming interface (API) unit 665, input unit 670,
output unit 675, and inter-unit communication mechanism 695 for the
different units to communicate with each other, with the OS, and
with other applications (not shown). The described units and
elements can be varied in design, function, configuration, or
implementation and are not limited to the descriptions
provided.
[0093] In an example implementation of a management system 220,
external storage 645 may be configured to store the management
information as illustrated in FIGS. 3(a) and 3(b), and processor(s)
610 may be configured to execute the functions as depicted in FIGS.
4(a) to 4(d).
[0094] In some example implementations, when information or an
execution instruction is received by API unit 665, it may be
communicated to one or more other units (e.g., logic unit 660,
input unit 670, and output unit 675.
[0095] In some instances, logic unit 660 may be configured to
control the information flow among the units and direct the
services provided by API unit 665, input unit 670, and output unit
675, in some example implementations described above. For example,
the flow of one or more processes or implementations may be
controlled by logic unit 660 alone or in conjunction with API unit
665.
[0096] Any of the software components described herein may take a
variety of forms. For example, a component may be a stand-alone
software package, or it may be a software package incorporated as a
"tool" in a larger software product. It may be downloadable from a
network, for example, a website, as a stand-alone product or as an
add-in package for installation in an existing software
application. It may also be available as a client-server software
application, as a web-enabled software application, and/or as a
mobile application.
[0097] Although a few example implementations have been shown and
described, these example implementations are provided to convey the
subject matter described herein to people who are familiar with
this field. It should be understood that the subject matter
described herein may be implemented in various forms without being
limited to the described example implementations. For example, but
not by way of limitation, the foregoing equations, expressions, and
numerical examples are for illustrative purposes, and others may be
substituted therefor, as would be understood by those skilled in
the art.
[0098] The subject matter described herein can be practiced without
those specifically defined or described matters or with other or
different elements or matters not described. It will be appreciated
by those familiar with this field that changes may be made in these
example implementations without departing from the subject matter
described herein as defined in the appended claims and their
equivalents.
* * * * *